Telemetry Reliability Overhaul

What: Failed telemetry events are now queued, batched, and retried with exponential backoff instead of being lost.

Details:

• Events saved to ~/.claude/telemetry/1p_failed_events.{session}.{batch}.json
• Maximum 8192 queued events, batch size of 200
• Exponential backoff: 500ms → 2s → 4.5s → 8s → 30s (capped)
• Events persist across CLI restarts and are retried on next session
• Default timeout increased from 5s to 10s
• Evidence: Class zm1 at line 234387, queueFailedEvents() at line 234542, retryPreviousBatches() at line 234448